Adaptation to prolonged neuromodulation in cortical ... - Noam Ziv Lab

Kaufman et al. BMC Biology 2014, 12:83 http://www.biomedcentral.com/1741-7007/12/83

RESEARCH ARTICLE

Open Access

Adaptation to prolonged neuromodulation in cortical cultures: an invariable return to network synchrony Maya Kaufman*, Sebastian Reinartz and Noam E Ziv

Abstract Background: Prolonged neuromodulatory regimes, such as those critically involved in promoting arousal and suppressing sleep-associated synchronous activity patterns, might be expected to trigger adaptation processes and, consequently, a decline in neuromodulator-driven effects. This possibility, however, has rarely been addressed. Results: Using networks of cultured cortical neurons, acetylcholine microinjections and a novel closed-loop ‘synchrony-clamp’ system, we found that acetylcholine pulses strongly suppressed network synchrony. Over the course of many hours, however, synchrony invariably reemerged, even when feedback was used to compensate for declining cholinergic efficacy. Network synchrony also reemerged following its initial suppression by noradrenaline, but this did not occlude the suppression of synchrony or its gradual reemergence following subsequent cholinergic input. Importantly, cholinergic efficacy could be restored and preserved over extended time scales by periodically withdrawing cholinergic input. Conclusions: These findings indicate that the capacity of neuromodulators to suppress network synchrony is constrained by slow-acting, reactive processes. A multiplicity of neuromodulators and ultimately neuromodulator withdrawal periods might thus be necessary to cope with an inevitable reemergence of network synchrony. Keywords: Neuromodulators, Synchrony, Acetylcholine, Closed-loop, Adaptation, Cultured neuronal networks, Multielectrode arrays

Background Neuronal networks are strongly influenced by neuromodulatory systems; consequently, the properties and dynamics of particular networks can vary enormously depending on the levels, timing and composition of neuromodulatory influences [1]. While such influences often vary on short time scales, neurons are also influenced by changes in neuromodulatory input over relatively long time scales. In many biological systems, prolonged exposure to agonists is associated with adaptation to those substances or their effects. Neurons, in particular, are known to adapt or react homeostatically to changes in their input levels or in their milieu (reviewed in [2-4]; see also [5,6]), raising the possibility that prolonged neuromodulation will be associated with some recovery of the affected properties. At present, * Correspondence: [email protected] Department of Physiology and Biophysics and Rappaport Institute, Technion Faculty of Medicine, and Network Biology Research Laboratories, Lorry Lokey Center for Life Sciences & Engineering, Fishbach Building, Haifa 32000, Israel

however, the question of adaptive or homeostatic reactivity to long-term neuromodulation has rarely been addressed. Some of the most important long-term neuromodulatory processes in the mammalian brain are those that regulate a striking form of cortical synchrony known as ‘slow oscillations’, ‘slow wave’ or ‘slow rhythmic’ activity. This activity pattern is characterized by transitions between periods of neuronal discharges (‘on’ periods) and periods of near-complete quiescence (‘off’ periods) which occur in remarkably synchronous fashion in large cortical domains. Macroscopically, these synchronous transitions appear in electroencephalogram (EEG) recordings as low frequency, high amplitude waves (for example, [7-15]; reviewed in [16]). At the single neuron level, these transitions typically coincide with transitions between relatively depolarized membrane potentials (‘up’ states), which are often accompanied by action potential firing and strong hyperpolarizations (‘down’ states) during which neurons are silent. Following the lead of Harris and Thiele [17] we

© 2014 Kaufman et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.


will refer to these network-wide, synchronous transitions between ‘on’ and ‘off’ periods as ‘synchrony’. Note that in this context synchrony does not refer to the degree to which multiple neurons fire action potentials simultaneously at millisecond time precision. In the intact brain, synchrony as defined above is strongly regulated by brainstem and basal forebrain noradrenergic and cholinergic neurons which project to widespread cortical regions [16,18-20]. The activation of these neuromodulatory systems strongly suppresses network synchrony and promotes asynchronous activity patterns typical of aroused and attentive behavioral states. In contrast, reduced activity of these systems, which occurs mainly during periods of NREM (non rapid eye movement) sleep, is associated with the prominent appearance of network synchrony as defined above. Importantly, however, neither these forms of network synchrony nor their modulation by acetylcholine (ACh) and noradrenaline (NA) are limited to the intact cortex, as similar activity patterns occur in brain slabs [21], acute and organotypic cortical preparations (for example, [22-26]) and even in networks of dissociated cortical neurons (for example, [27-37]; reviewed in [38]). Where networks of dissociated cortical neurons in culture are concerned, synchrony takes the form of networkwide bursting activity which lasts for several hundreds of milliseconds, separated by longer periods (1 to 10 seconds) of near-complete quiescence or sparse, asynchronous action potentials [27-37]. These network-wide bursts are less frequent and more stereotyped as compared to those observed in the intact brain, which might be expected given the smaller size and lower connection density of these networks as well as the lack of reentrant pathways [21,39-41]). Moreover, it has been suggested that the degree of synchrony in these and other in vitro preparations is exacerbated by various homeostatic responses to deafferentation, resulting in activity forms that share some similarities with seizure-related paroxysmal activity (as indicated by in vivo deafferentation studies [42,43]). Yet, while the forms of synchrony observed in vitro differ in many respects from those associated with low arousal levels in the intact brain, their underlying biophysical mechanisms share important similarities. Both in vivo [10,15,17,21,44] and in vitro [27,33,34,37,45-47] experiments, as well as modeling studies [21,39,40,48,49], indicate that these forms of synchrony are not imposed by some external circuitry, global inhibition or pacemaker cells, but probably arise from the interplay of spontaneous synaptic activity, nonlinear neuronal recruitment cascades, refractoriness and network wide synaptic depression (summarized in [17]), effectively giving rise to a default activity mode, as it has been referred to [35,36] (see also [39]). Furthermore, and in full concordance with their activities in vivo [44,50-61], cholinergic and adrenergic

Page 2 of 22

agonists suppress network synchrony in cell culture and slice preparations, shifting spontaneous activity away from this ‘default’ mode towards desynchronized, tonic firing modes [35,36,62-65]. Thus, while synchrony in networks of cultured cortical neurons does not fully replicate the forms of synchrony related to low neuromodulatory tone in the intact brain, the similarities in underlying mechanisms and the comparable effects of neuromodulation suggest that this preparation is a useful model system for studying relationships between prolonged neuromodulation and network synchrony. As mentioned above, the tendency of neurons and neuronal networks to adapt or react homeostatically raises the possibility that prolonged neuromodulation will be associated with some reactive adaptation over long time scales. Indeed, in a prior study [35] we found that the chronic, prolonged (many hour) exposure of networks of cultured cortical neurons to a cholinergic agonist is associated with the gradual growth of excitatory synapses and, intriguingly, to the gradual reemergence of synchrony (see also [36]). If neuronal networks adapt to neuromodulatory input, it might be asked how the necessity to suppress network synchrony is ultimately addressed, in particular given that this activity form seems to be incompatible with attentive states and arousal [66,67]. Conceivably, the contradiction between this necessity and diminishing neuromodulator efficacy might have been resolved by adjusting neuromodulatory input to match instantaneous neuromodulator efficacy. Alternatively, this contradiction might have been alleviated by the existence of multiple neuromodulatory systems [18] that exert similar effects but employ different cellular mechanisms. If, however, neither of these routes resolve the need to suppress synchrony, periodic neuromodulator withdrawal periods (such as those which occur during NREM sleep periods) might be ultimately required. To date, however, none of these possibilities have been explored or addressed experimentally. Here we used a system based on networks of cultured cortical neurons, ACh microinjections and a novel closedloop ‘synchrony-clamp’ to address these questions. Specifically, we examined the capacity of both fixed and feedback-based adjusted cholinergic input to suppress network synchrony continually as defined above. We then examined the ability of multiple neuromodulators to suppress network synchrony continually. Finally we examined the possibility that cholinergic neuromodulatory efficacy might be preserved on extended time scales by periodically withdrawing cholinergic input.

Results Rational and experimental approach

To examine relationships between prolonged cholinergic input and network synchrony we developed an experimental


system which allowed us to tightly control and manipulate cholinergic input, measure its effects on network activity and synchrony, and assay changes in its capacity to suppress synchrony, with the latter serving as a measure of adaptive or homeostatic reactions occurring in the same networks. This system, shown schematically in Figure 1, is based on primary cultures of dissociated rat cortical neurons growing on substrate integrated multielectrode arrays (MEAs), a microinjection system used to apply minute ACh solution volumes in the form of brief pulses, and a controller used to apply ACh at fixed intervals (open-loop experiments) or at intervals modified online as required to ‘clamp’ synchrony at fixed, low levels (closed-loop experiments). To terminate ACh signaling after each application and avoid ACh buildup in the media, acetylcholine esterase (AChE) was added to the network cell culture media (and perfusion system), assuring continuous and efficient ACh enzymatic breakdown [68]. ACh injections were delivered via a needle immersed in the culture media, hovering about 1 mm above the neuronal network (Figure 1a, item 6), whereas a slow continuous mixing system was used to accelerate the dilution and distribution of the applied ACh and facilitate its subsequent enzymatic breakdown (Figure 1a, item 7). In addition, the system also included provisions for optimal environmental conditions (a slow perfusion system, a stream of a sterile gas mixture and heating devices, Figure 1a, items 3 to 5), providing a stable environment of 37°C, 5% CO2 and one to two media replacements/day, resulting in experiments which were effectively open-ended [69]. Pulsed ACh applications effectively suppress network synchrony, but synchrony eventually reemerges

Under baseline conditions, and in agreement with many reports [27,30-36,70], spontaneous activity in the cortical networks used here occurs as periods of synchronous, network-wide bursting activity which lasts for several hundreds of milliseconds, separated by longer periods (1 to 10 seconds) of near-complete quiescence or sparse, asynchronous action potentials (Figure 2a). As mentioned in the introduction, these network activity patterns share many similarities with the forms of synchrony observed in vivo under regimes of low neuromodulatory levels. As shown in Figure 2b, this similarity also extends to the single neuron level: intracellular whole-cell recordings from individual neurons (six neurons in three networks) performed concomitantly with extracellular recordings from the 59 electrodes of the MEAs invariably showed that periods of network-wide bursting activity were tightly correlated with neuronal membrane potential depolarizations, often accompanied by the firing of action potentials, whereas network quiescence periods were associated with deep and constant membrane hyperpolarizations (Figure 2a,b). Closer examination (Figure 2d) revealed that these depolarizations and hyperpolarizations

Page 3 of 22

exhibited a marked resemblance to ‘up’ and ‘down’ states recorded in vivo (compare, for example, with [71]), as did their tight temporal correlation with network ‘on’ and ‘off’ periods (Figure 2c), although their occurrence was less frequent (see Background). Our first goal was to examine how the forms of network synchrony described above are affected by cholinergic input delivered at a fixed rate. To that end, we performed the following experiments: cortical networks growing on MEA dishes (maintained in culture for at least 17 days) were mounted on the MEA headstage and provided with optimal environmental conditions as described above. After several preparatory phases (Figure 1c, phases I-IV; see Additional file 1: Figure S1 and Methods), including the activation of the slow continuous mixing system mentioned above (Figure 1a, item 7), the experiments were started by initiating automated ACh applications at fixed intervals (Figure 1c, phase V), that is, in an ‘open-loop’ regime as illustrated in Figure 1b. ACh applications consisted of minute volumes (1 μl) of concentrated ACh solution (20 mM) briefly injected at five minute intervals into the media bathing the cortical networks, which, as mentioned above, contained AChE and was continually mixed. Ignoring ACh breakdown, a single, well-mixed ACh application would be expected to elevate ACh concentration to a final value of 10 μM. The particular experimental profile used here was chosen after an extensive series of preliminary experiments in which we explored a wide range of ACh concentrations and application rates, settling on this regime as a compromise between the desire to mimic physiological profiles of cholinergic neuromodulation and the constraints imposed by the finite volumes of the injection syringe and MEA dish, mixing rates, synchrony stability and the time course of synchrony recovery. One fixed input rate experiment is shown in Figure 3a,b. As shown here, prior to ACh applications synchrony levels were high and quite stable (see also Additional file 1: Figure S1 and Additional file 2: Figure S2). The initiation of ACh applications (starting exactly five minutes after the beginning of the recording phase shown) had major effects on the characteristics of spontaneous activity, the most obvious being the replacement of stereotyped, networkwide alternating periods of activity and silence with much more diverse activity forms (compare the raster plots of Figure 2a with Figure 3a and Additional file 1: Figure S1 and Additional file 3: Figure S3). These ranged from near continuous tonic firing, through intermittent tonic firing, through sporadic firing to near silence. Importantly, and in excellent agreement with prior in vitro [35,62-65] and in vivo [50,57,61] studies, the network-wide synchrony of neural activity was strongly reduced. To quantify the effects of ACh on network synchrony we developed a robust measure that was termed the ‘Sync Ratio’ which ranges


a

Page 4 of 22

7

6

5

2 1

8 3 9

4 1. 2. 3. 4.

5. 6. 7. 8. 9.

Network of cortical neurons growing on MEA dish MEA headstage Ultraslow peristaltic perfusion system Reservoir containing cell culture medium + AChE

Airflow regulator (5% CO2, 95% air mixture) ACh injection system (Ultra MicroPump) Continuous mixing system Closed loop controller (Simulink xPC target) User interface computer (Windows, Matlab)

b Predetermined Application Times

Application system

Open Loop

Cortical network on MEA 59 Channel Data acquisition , Spike detection & Data storage

Closed Loop

c

Application system

I Slow perfusion (24-72 h)

Figure 1 (See legend on next page.)

PI controller: Calculate IAI

II III IV Mixing AChE Needle (12-24 h) added insertion

Calculate Sync Ratio + Error

V ACh applications


Page 5 of 22

(See figure on previous page.) Figure 1 Setup used to examine long-term effects of ACh applications on network synchrony in open-and closed-loop regimes. a) Schematic illustration of the experimental system used to record from neurons growing on MEA substrates, maintain their viability, and apply ACh pulses at predetermined rates or at rates adjusted online to maintain network synchrony at predefined levels. b) Schematic illustration of open- and closed-loop experimental protocols. Note that in open-loop experiments ACh application timings are predefined, whereas in closed-loop experiments, ACh application timings are determined online according to instantaneous Sync Ratio values calculated by the real-time controller (item 8 in a, see Additional file 3: Figure S3 for an explanation of the Sync Ratio measure). c) Preparatory and experimental phases. Each experiment was preceded by the following preparatory phases: Phase I, slow perfusion (at least 24 hours); Phase II, activation of continuous mixing system (12 to 24 hours); Phase III, addition of AChE (0.1 U/ml) to the MEA dish and perfusion media reservoir (30 minutes, see Additional file 7: Figure S6 for the importance of AChE addition); Phase IV, insertion of ACh application needle into the MEA dish; and Phase V, experiment (open or closed loop). Note that the experimental component added in each phase was in effect from that moment until the end of the experiment. See Methods for further details. ACh, acetylcholine; AChE, acetylcholine esterase; MEA, multielectrode array.

from 0 to 1 (completely asynchronous to completely synchronous activity, respectively; see Methods and Additional file 3: Figure S3 for a full explanation of this measure’s rational and derivation). In this experiment, the Sync Ratio was reduced from approximately 0.9 (mixing system active, no exposure to ACh; Figure 1c, phase II) to approximately 0.2. In general, each ACh application was associated with a transient decrease in the Sync Ratio which then partially recovered over the next few minutes (Figure 3b). As applications were delivered at five minute intervals, network synchrony did not have sufficient time to recover fully. Importantly, after about 10 hours, Sync Ratio values started to gradually increase, finally settling on an intermediate Sync Ratio value after about 20 hours (Figure 3a, note also right-hand raster plot). This reemergence of synchrony occurred in spite of the fact that almost every ACh application was still partially effective as shown in Figure 3b. The average evolution of the Sync Ratio (normalized to the average Sync Ratio during the four-hour pre ACh applications period) for five fixed input rate (open-loop) experiments (performed in five separate preparations) is shown in Figure 3c. As shown here, prolonged ACh applications at five-minute intervals were followed by a significant, yet incomplete, recovery of network synchrony to approximately 65% of initial Sync Ratio levels. Although absolute Sync Ratio values varied slightly between experiments (see Additional file 2: Figure S2a), all experiments showed a qualitatively similar, partial but significant recovery of network synchrony (as illustrated in detail for one experiment in Additional file 4: Video S1). The time to recovery in these experiments (defined as illustrated in Additional file 5: Figure S4a) ranged from 10 to 16 hours (see Additional file 5: Figure S4b-d). No clear relationships were found between the time to recovery and baseline synchrony or burst characteristics (see Additional file 5: Figure S4b-d). The initial (approximately 10 hours) periods of suppressed network synchrony were not accompanied by reductions in mean firing rates as compared to pre-ACh

applications levels (Figure 3d), indicating that Sync Ratio values can change independently of changes in firing rates (compare Figure 3c and d). On the other hand, synchrony reemergence was associated with some increase in firing rates (Figure 3d). This increase was accompanied by elevated burst rates, (see Additional file 6: Figure S5a), but burst intensity remained smaller than that observed before ACh was applied (see Additional file 6: Figure S5b; see also Additional file 4: Video S1), indicating that the increase in firing rates was probably not entirely attributable to excess bursting. As mentioned above, on a time-scale of minutes, each ACh application caused a rapid reduction of Sync Ratio values followed by a partial recovery over the next few minutes. These temporal profiles were probably dictated by a combination of network responses to changes in cholinergic input, ACh mixing kinetics, ACh breakdown kinetics, and, possibly, the kinetics of potential adaptive processes. We noted that in the absence of AChE, practically no recovery of the Sync Ratio was observed for >2 hours following a single ACh application (see Additional file 7: Figure S6). Conversely, 10-fold increases in AChE concentrations accelerated Sync Ratio recovery rates only by a factor of 10 hours,

Page 10 of 22

respectively. These experiments indicate that changes in firing rates induced by various neuromodulators are not necessarily associated with changes in network synchrony and highlight the specific effects of ACh and NA in this respect. They also indicate that prolonged exposure to a third neuromodulator – DA – is also followed by adaptive or reactive processes which occur over many hours, in line with our observations for ACh and NA. Periodic ‘withdrawal’ periods restore the capacity of ACh applications to suppress network synchrony in ensuing periods

As mentioned in the introduction, timed cholinergic withdrawal periods (such as those which occur during NREM sleep periods) might ultimately be required to preserve cholinergic efficacy over extended time scales. In the setting of our experimental system, this predicts that withdrawal periods introduced into our experimental protocols will restore the capacity of ACh to suppress network synchrony in subsequent periods. To test this prediction, we performed experiments that included three closed-loop epochs, during which network synchrony was ‘clamped’ at low Sync Ratio values, separated by periods during which no ACh was applied (‘withdrawal’ periods). The duration of these periods was chosen to be 12 hours, to mirror the times at which synchrony often started to reemerge following prolonged cholinergic input (Figure 3). As shown in Figure 6a, multiple closed-loop epochs were qualitatively similar to the closed-loop experiments described above. In particular, as in the aforementioned experiments, the demand to maintain synchrony at low predetermined levels was associated with gradual and very significant decreases in IAIs over time. During ‘withdrawal’ periods, synchrony returned almost immediately to the high levels typical of unperturbed networks, as manifested by Sync Ratio values that were similar or even exceeded initial values. Notably, ACh applications delivered after ‘withdrawal’ periods were highly effective in desynchronizing network activity. Moreover, gauging the degree of adaptation from the IAIs needed to clamp network synchrony at the beginning of the second and third closedloop epochs and comparing these to IAI values at the end of the preceding epochs, indicated that cholinergic efficacy had recovered and that the degree of adaptation to cholinergic input was low. In the experiment presented in Figure 6a, the ‘withdrawal’ phase of each epoch was started when the ACh application syringe was depleted. Accordingly, the application syringe had to be refilled between epochs, a process that was associated with a small leak of ACh into the culture media (Figure 6a). Moreover, the duration of the closed-loop epochs in this experiment became gradually shorter, reflecting a diminished capability


a

1

Page 11 of 22

NA (20µM)

NA (20µM)

10

20

CCh (20µM)

0.9 0.8 Sync Ratio

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0

30

40 Time (hours)

50

60

70

b

c

NA (20µM)

1

CCh (20µM)

0.9

Experiment 1 Experiment 2 Experiment 3

0.8 Sync Ratio

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0

10

20

30

40

50

Time (hours)

Figure 5 Network synchrony reemerges following prolonged exposure to NA, but can be re-suppressed by cholinergic input. a) Evolution of the Sync Ratio in one experiment (grey; same data after smoothing with a five-point kernel is shown in black). After recording baseline activity for >72 hours (of which the last approximately six hours are shown), NA (20 μM) was added directly into the MEA dish and the perfusion media reservoir (see Additional file 11: Figure S8 for similar experiments performed with DA). Nineteen hours later, a second bolus of freshly prepared NA was added to the MEA dish. Twenty-four hours after the first NA addition, CCh (20 μM) was added to the MEA dish and perfusion media reservoir and recording was continued for 40 hours. b) Examples of one-minute raster plots from four stages of the experiment as indicated in the figure (green arrows). c) Evolution of Sync Ratio (averaged over 30 minute intervals) in three similar experiments (blue line is the same experiment as in a). In two of these, a second bolus of freshly prepared NA was added to the MEA dish 19 hours later (indicated as small circles on blue and red traces). Note the very limited effect of this second NA application. CCh, carbachol; DA, dopamine; MEA, multielectrode array; NA, noradrenaline.


Page 12 of 22

Figure 6 Withdrawal of cholinergic input for defined periods restores its capacity to suppress network synchrony in subsequent periods. Three multiple epoch experiments are shown using the same notations as in Figure 4. Network activity was clamped to a desynchronized activity level (Sync Ratio = 0.05 for a, b and 0.1 and 0.15 for c). Following the depletion of the ACh in the application syringe (a) or the online calculation of four consecutive IAI values